Not able to optimize query - mysql

Question

SELECT MIN(classification) AS classification
,MIN(START) AS START
,MAX(next_start) AS END
,SUM(duration) AS seconds
FROM ( SELECT *
, CASE WHEN (duration < 20*60) THEN CASE WHEN (duration = -1) THEN 'current_session' ELSE 'session' END
ELSE 'break'
END AS classification
, CASE WHEN (duration > 20*60) THEN ((#sum_grouping := #sum_grouping +2)-1)
ELSE #sum_grouping
END AS sum_grouping
FROM ( SELECT *
, CASE WHEN next_start IS NOT NULL THEN TIMESTAMPDIFF(SECOND, START, next_start) ELSE -1 END AS duration
FROM ( SELECT id, studentId, START
, (SELECT MIN(START)
FROM attempt AS sub
WHERE sub.studentId = main.studentId
AND sub.start > main.start
) AS next_start
FROM attempt AS main
WHERE main.studentId = 605
ORDER BY START
) AS t1
) AS t2
WHERE duration != 0
) AS t3
GROUP BY sum_grouping
ORDER BY START DESC, END DESC
Explanation and goal
The attempt table records a student's attempt at some activity, during a session. If two attempts are less than 20 minutes apart, we consider those to be the same session. If they are more than 20 minutes apart, we assume they took a break.
My goal with this query is to take all of the attempts and condense them down in a list of sessions and breaks, with the start time of each session, the end time (defined as the start of the subsequent session), and how long the session was. The classification is whether it is a session, a break, or the current session.
The above query does all of that, but is too slow. How can I improve the performance?
How the current query works
The innermost queries select an attempt's start time and the subsequent attempt's start time, along with the duration between those values.
Then, the #sum_grouping and sum_grouping are used to split the attempts into the sessions and breaks. #sum_grouping is only ever increased when an attempt is more than 20 minutes long (i.e. a break), and it is always increased by 2. However, sum_grouping is set to a value of one less than that for that "break". If an attempt is less than 20 minutes long, then the current #sum_grouping value is used, without modification. As a result, all breaks are distinct odd values, and all sessions (whether of 1 or more attempt) end up as distinct even numbers. This allows the GROUP BY portion to correctly separate the attempts into sessions and breaks.
Example:
Attempt type #sum_grouping sum_grouping
non-break 0 0
non-break 0 0
break 2 1
break 4 3
non-break 4 4
break 6 5
As you can see, all the breaks will be grouped by sum_grouping separately with distinct odd values and all the non-breaks will be grouped together as sessions with the even values.
The MIN(classification) simply forces "current session" to be returned when both "session" and "current session" are present within a grouped row.
OUTPUT OF SHOW CREATE TABLE attempt
CREATE TABLE attempt (
id int(11) NOT NULL AUTO_INCREMENT,
caseId int(11) NOT NULL DEFAULT '0',
eventId int(11) NOT NULL DEFAULT '0',
studentId int(11) NOT NULL DEFAULT '0',
activeUuid char(36) NOT NULL,
start timestamp NOT NULL DEFAULT '0000-00-00 00:00:00',
end timestamp NULL DEFAULT NULL,
outcome float DEFAULT NULL,
response varchar(5000) NOT NULL DEFAULT '',
PRIMARY KEY id),
KEY activeUuid activeUuid),
KEY caseId caseId,activeUuid),
KEY end end),
KEY start start),
KEY studentId studentId),
KEY attempt_idx_studentid_stat_id studentId,start,id),
KEY attempt_idx_studentid_stat studentId,start
) ENGINE=MyISAM AUTO_INCREMENT=298382 DEFAULT CHARSET=latin1

Answer 1

(This is not a proper Answer, but here goes anyway.)
Try not to nest 'derived' tables.
I see a lot of syntax errors.
Move from MyISAM to InnoDB.
INDEX(a, b) handles situations where you need INDEX(a), so DROP the latter.